Video Podcast Maker

# Learn from images (use your agent's image analysis capability to analyze design patterns)
python3 learn_design.py ./screenshot1.png ./screenshot2.png

# Learn from a local video (ffmpeg extracts frames automatically)
python3 learn_design.py ./reference.mp4

# Learn from a URL (Playwright captures screenshots — experimental)
python3 learn_design.py https://www.bilibili.com/video/BV1xx411c7mD

# Save with a named profile and tags
python3 learn_design.py ./reference.mp4 --profile "tech-minimal" --tags "tech,minimal,dark"

Reference Library Commands

references list          # List all stored references (auto-cleans orphaned entries)
references show <id>     # Show full design report for a reference
references delete <id>   # Delete a reference and its files

Style Profile Commands

profiles list            # List all saved style profiles
profiles show <name>     # Show profile props_override
profiles delete <name>   # Delete a style profile
profiles create <name>   # Create a new style profile interactively

Pre-Workflow Usage

When the user provides a reference video or image alongside a video creation request, extract design patterns before Step 1 and apply them as session overrides. See references/workflow-steps.md → Pre-workflow section for the full extraction flow.

Step 9 Integration

Before choosing visual design in Step 9, check for matching style profiles or reference library entries. Apply the best match as a starting point for Remotion composition props. See references/workflow-steps.md → Step 9 Style Profile Integration for the priority chain.

Auto Update Check

Agent behavior: Check for updates at most once per day (throttled by timestamp file). Before any shell command that reads files from this skill, resolve SKILL_DIR to the directory containing SKILL.md. If your agent exposes a built-in skill directory variable such as ${CLAUDE_SKILL_DIR}, you may map it to SKILL_DIR.

SKILL_DIR="${SKILL_DIR:-${CLAUDE_SKILL_DIR}}"
STAMP="${SKILL_DIR}/.last_update_check"
NOW=$(date +%s)
LAST=$(cat "$STAMP" 2>/dev/null || echo 0)
if [ $((NOW - LAST)) -gt 86400 ]; then
  timeout 5 git -C "${SKILL_DIR}" fetch --quiet 2>/dev/null || true
  LOCAL=$(git -C "${SKILL_DIR}" rev-parse HEAD 2>/dev/null)
  REMOTE=$(git -C "${SKILL_DIR}" rev-parse origin/main 2>/dev/null)
  echo "$NOW" > "$STAMP"
  if [ -n "$LOCAL" ] && [ -n "$REMOTE" ] && [ "$LOCAL" != "$REMOTE" ]; then
    echo "UPDATE_AVAILABLE"
  else
    echo "UP_TO_DATE"
  fi
else
  echo "SKIPPED_RECENT_CHECK"
fi

• Update available: Ask the user whether to pull updates. Yes → git -C "${SKILL_DIR}" pull. No → continue.
• Up to date / Skipped: Continue silently.

Prerequisites Check

!( missing=""; node -v >/dev/null 2>&1 || missing="$missing node"; python3 --version >/dev/null 2>&1 || missing="$missing python3"; ffmpeg -version >/dev/null 2>&1 || missing="$missing ffmpeg"; [ -n "$AZURE_SPEECH_KEY" ] || missing="$missing AZURE_SPEECH_KEY"; if [ -n "$missing" ]; then echo "MISSING:$missing"; else echo "ALL_OK"; fi )

If MISSING reported above, see README.md for full setup instructions (install commands, API key setup, Remotion project init).

Overview

Automated pipeline for professional Bilibili horizontal knowledge videos from a topic.

Target: Bilibili horizontal video (16:9)

• Resolution: 3840×2160 (4K) or 1920×1080 (1080p)
• Style: Clean white (default)

Tech stack: Coding agent + TTS backend + Remotion + FFmpeg

Output Specs

Execution Modes

Agent behavior: Detect user intent at workflow start:

• "Make a video about..." / no special instructions → Auto Mode
• "I want to control each step" / mentions interactive → Interactive Mode
• Default: Auto Mode

Auto Mode (Default)

Full pipeline with sensible defaults. Mandatory stop at Step 9:

1. Step 9: Launch Remotion Studio — user reviews in real-time, requests changes until satisfied
2. Step 10: Only triggered when user explicitly says "render 4K" / "render final version"

Users can override any default in their initial request:

• "make a video about AI, burn subtitles" → auto + subtitles on
• "use dark theme, AI thumbnails" → auto + dark + imagen
• "need screenshots" → auto + media collection enabled

Interactive Mode

Prompts at each decision point. Activated by:

• "interactive mode" / "I want to choose each option"
• User explicitly requests control

Workflow State & Resume

Planned feature (not yet implemented). Currently, workflow progress is tracked via the agent's conversation context. If a session is interrupted, re-invoke the skill and inspect existing files in videos/{name}/ to determine where to resume.

Technical Rules

Hard constraints for video production. Visual design remains the agent's creative freedom within these rules:

Additional Resources

Load these files on demand — do NOT load all at once:

• references/workflow-steps.md: Detailed step-by-step instructions (Steps 1-14). Load at workflow start.
• references/design-guide.md: Visual minimums, typography, layout patterns, checklists. MUST load before Step 9.
• references/troubleshooting.md: Error fixes, BGM options, preference commands, preference learning. Load on error or user request.
• examples/: Real production video projects. The agent may reference these for composition structure and timing.json format.

Directory Structure

project-root/                           # Remotion project root
├── src/remotion/                       # Remotion source
│   ├── compositions/                   # Video composition definitions
│   ├── Root.tsx                        # Remotion entry
│   └── index.ts                        # Exports
│
├── public/                             # Remotion default (unused — use --public-dir videos/{name}/)
│
├── videos/{video-name}/                # Video project assets
│   ├── workflow_state.json             # Workflow progress
│   ├── topic_definition.md             # Step 1
│   ├── topic_research.md               # Step 2
│   ├── podcast.txt                     # Step 4: narration script
│   ├── podcast_audio.wav               # Step 8: TTS audio
│   ├── podcast_audio.srt               # Step 8: subtitles
│   ├── timing.json                     # Step 8: timeline
│   ├── thumbnail_*.png                 # Step 7
│   ├── output.mp4                      # Step 10
│   ├── video_with_bgm.mp4             # Step 11
│   ├── final_video.mp4                 # Step 12: final output
│   └── bgm.mp3                         # Background music
│
└── remotion.config.ts

Important: Always use --public-dir and full output path for Remotion render:

npx remotion render src/remotion/index.ts CompositionId videos/{name}/output.mp4 --public-dir videos/{name}/

Naming Rules

Video name {video-name}: lowercase English, hyphen-separated (e.g., reference-manager-comparison)

Section name {section}: lowercase English, underscore-separated, matches [SECTION:xxx]

Thumbnail naming (16:9 AND 4:3 both required):

Public Directory

Use --public-dir videos/{name}/ for all Remotion commands. Each video's assets (timing.json, podcast_audio.wav, bgm.mp3) stay in its own directory — no copying to public/ needed. This enables parallel renders of different videos.

# All render/studio/still commands use --public-dir
npx remotion studio src/remotion/index.ts --public-dir videos/{name}/
npx remotion render src/remotion/index.ts CompositionId videos/{name}/output.mp4 --public-dir videos/{name}/ --video-bitrate 16M
npx remotion still src/remotion/index.ts Thumbnail16x9 videos/{name}/thumbnail.png --public-dir videos/{name}/

Workflow

Progress Tracking

At Step 1 start:

1. Create videos/{name}/workflow_state.json
2. Use TaskCreate to create tasks per step. Mark in_progress on start, completed on finish.
3. Each step updates BOTH workflow_state.json AND TaskUpdate.

 1. Define topic direction → topic_definition.md
 2. Research topic → topic_research.md
 3. Design video sections (5-7 chapters)
 4. Write narration script → podcast.txt
 5. Collect media assets → media_manifest.json
 6. Generate publish info (Part 1) → publish_info.md
 7. Generate thumbnails (16:9 + 4:3) → thumbnail_*.png
 8. Generate TTS audio → podcast_audio.wav, timing.json
 9. Create Remotion composition + Studio preview (mandatory stop)
10. Render 4K video (only on user request) → output.mp4
11. Mix background music → video_with_bgm.mp4
12. Add subtitles (optional) → final_video.mp4
13. Complete publish info (Part 2) → chapter timestamps
14. Verify output & cleanup
15. Generate vertical shorts (optional) → shorts/

Validation Checkpoints

After Step 8 (TTS):

• podcast_audio.wav exists and plays correctly
• timing.json has all sections with correct timestamps
• podcast_audio.srt encoding is UTF-8

After Step 10 (Render):

• output.mp4 resolution is 3840x2160
• Audio-video sync verified
• No black frames

Key Commands Reference

See CLAUDE.md for the full command reference (TTS, Remotion, FFmpeg, shorts generation).

User Preference System

Skill learns and applies preferences automatically. See references/troubleshooting.md for commands and learning details.

Storage Files

Priority

Final = merge(Root.tsx defaults < global < topic_patterns[type] < current instructions)

User Commands

Troubleshooting & Preferences

Full reference: Read references/troubleshooting.md on errors, preference questions, or BGM options.